Motorbike Counting in Heavily Crowded Scenes

Vehicle density estimation has an important role in intelligent traffic systems. As of now, most established studies only focused on areas where people mainly travel by four-wheeled vehicles rather than motorbikes. However, in some countries such as Vietnam where motorbikes are the majority, traffic scenarios will pose different issues. Motorbikes are intrinsically more flexible so they can cause cluttered and chaotic visual. As a result, traffic video data captured in such environment is more challenging to existing systems. In this work, we performed an empirical survey on a set of vision-based counting methods covering a wide range of models and techniques. To our knowledge, there has not been many works dedicated to tackle this problem. Based on our experimental results, some of the top performers is ready to be used in real systems due to their robustness.